Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Delay Aware Packet Scheduling (DAPS) and receiver’s buffer blocking in CMT-SCTP Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA (Toulouse, France) 2 National ICT Australia nicolas.kuhn@isae.fr July 31, 2013 Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Context and problem definition Context: many recent studies on ways to adapt TCP to the multi-path data transfert; we focus on Concurrent Multipath Transfer Using SCTP Multihoming (solution proposed may be transfered to MP-TCP); asymmetric heterogeneous network conditions: receiver’s buffer blocking issues (out of order packet arrival). Content of the presentation: measure the receiver’s buffer blocking time; propose a solution (adaptation of CMT-SCTP scheduler) for that specific issue; compare CMT-SCTP with and without our scheduler, Delay Aware Packet Scheduler. Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Overview 1 Window blocking issues and maximum blocking time 2 Delay Aware Packet Scheduling 3 Conclusion Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Maximum blocking time with CMT-SCTP Events at t 0: (1) t0 TSN (2) . . . TSN ( N chunk ) on RTTm/2 path mRTT - (2) TSN (1) on 2 t1 3 path MRTT ; 4 5 at t 0: RWND : block state; at t 1: reception of TSN (2); RTTM/2 at t 2: reception of TSN (1) - packet forwarded to app.; 1 T maxblock = t 2 − t 1 T maxblock = RTT MRTT L ∗ 8 + t2 C MRTT − 2 RTT mRTT L ∗ 8 − 2 C mRTT Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Validation blocking time Worst case model NS-2 C0 C1 RTT1 T maxblock (Mb/s) (Mb/s) (ms) Model 0.1 10 10 100 0.040 0.09 Blocking time (s) 0.08 50 10 100 0.041 0.07 0.06 10 50 100 0.039 0.05 100 100 100 0.040 0.04 0.03 50 150 100 0.039 0.02 0.01 150 50 100 0.040 0 150 160 170 180 190 200 10 10 200 0.090 Packet ID 50 10 200 0.091 C0 100Mbps, RTT1 100ms,C1 100Mbps (NS−2) C0 100Mbps, RTT1 100ms,C1 100Mbps (MODEL) 10 50 200 0.089 C0 100Mbps, RTT1 200ms,C1 100Mbps (NS−2) C0 100Mbps, RTT1 200ms,C1 100Mbps (MODEL) 100 100 200 0.090 C0 10Mbps, RTT1 200ms,C1 10Mbps (NS−2) C0 10Mbps, RTT1 200ms,C1 10Mbps (MODEL) 50 150 200 0.089 150 50 200 0.090 We measure an important blocking time in the context of asymmetric links. We propose to adapt the scheduling in CMT to reduce this delay which may “block” the receiver’s buffer. Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Overview 1 Window blocking issues and maximum blocking time 2 Delay Aware Packet Scheduling 3 Conclusion Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Delay Aware Packet Scheduling CMT DAPS RWND=10 RWND=10 on path 1 on path 1 3 on path 2 1 on path 2 4 2 5 3 6 4 DAPS: 7 5 8 compute {14,15} 6 9 to send on low path 7 10 for in sequence delivery BLOCK! 8 9 1 10 11 2 12 13 deliver 1 to app 14 deliver 2, ..., 10 to app 15 Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion DAPS performance Parameters RWND = 655 kB ; L = 1500 B ; Butterfly RTT 0 = 20 ms ; C 0 = [1 Mbps ; 1 . 5 Mbps ]; RTT0, C0 1 4 RTT 1 ∈ [100; 200] ms ; RWND C 1 ∈ [500 kbps ; 1 Mbps ] 0 3 Asymmetry is defined as: A = ( C 1 . RTT 1) / ( C 0 . RTT 0) RTT1, C1 5 Case C0 C1 RTT1 A 2 (Mb/s) (Mb/s) (ms) 1 1 1 100 5 2 1 1 200 10 3 1.5 0.5 200 3.33 Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Asymmetry: 5 CMT Case 1 DAPS Case 1 1800 1600 1800 1400 1600 1200 1400 Packet ID 1000 1200 Packet ID 800 1000 600 800 400 600 200 400 0 200 0 2 4 6 8 10 12 0 Simulation time (s) 0 2 4 6 8 10 12 Packet in Sending Buffer Packet received in order Simulation time (s) Packet in Sending Buffer Packet received in order 0.8 0.7 0.8 0.6 0.7 Blocking time (s) 0.5 0.6 Blocking time (s) 0.4 0.5 0.3 0.4 0.3 0.2 0.2 0.1 0.1 0 0 200 400 600 800 1000 1200 1400 1600 1800 0 Packet ID 0 200 400 600 800 1000 1200 1400 1600 1800 Packet ID Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Asymmetry: 10 CMT Case 2 DAPS Case 2 1600 1400 1800 1200 1600 1400 1000 Packet ID 1200 800 Packet ID 1000 600 800 400 600 200 400 0 200 0 1 2 3 4 5 6 7 8 9 10 0 Simulation time (s) 0 2 4 6 8 10 12 Packet in Sending Buffer Packet received in order Simulation time (s) Packet in Sending Buffer Packet received in order 0.7 0.6 0.6 0.5 0.5 Blocking time (s) 0.4 Blocking time (s) 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0 0 200 400 600 800 1000 1200 1400 1600 0 Packet ID 0 200 400 600 800 1000 1200 1400 1600 Packet ID Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Asymmetry: 3.33 CMT Case 3 DAPS Case 3 1800 1600 1800 1400 1600 1200 1400 Packet ID 1000 1200 Packet ID 800 1000 600 800 400 600 200 400 0 200 0 2 4 6 8 10 12 0 Simulation time (s) 0 2 4 6 8 10 12 Packet in Sending Buffer Packet received in order Simulation time (s) Packet in Sending Buffer Packet received in order 1.4 1.4 1.2 1.2 1 Blocking time (s) 1 0.8 Blocking time (s) 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 200 400 600 800 1000 1200 1400 1600 0 Packet ID 0 200 400 600 800 1000 1200 1400 1600 Packet ID Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion Overview 1 Window blocking issues and maximum blocking time 2 Delay Aware Packet Scheduling 3 Conclusion Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Window blocking issues and maximum blocking time Delay Aware Packet Scheduling Conclusion DAPS Performance: first conclusions In case 2 (Asymmetry: 10), DAPS provide serious improvements; In case 1 (Asymmetry: 5), improvements are limited (without doing worse than CMT-SCTP); In case 3 (Asymmetry: 3.33), improvements are limited. The performance of DAPS are related to: the Asymmetry of the links; the receiver’s buffer, which we will later investigated on to improve the algorithm. Nicolas KUHN 1 , 2 , Golam SARWAR 2 , Emmanuel LOCHIN 1 , Roksana BORELI 2 , Ahlem MIFDAOUI 1 1 ISAE / T´ eSA, 2 NICTA Delay Aware Packet Scheduling (DAPS)
Recommend
More recommend
